EXPLANATION THROUGH DIALOGUE FOR RULE-BASED REASONING AI SYSTEMS

Abstract

Explainable Artificial Intelligence (XAI) systems have garnered increasing attention across various domains, such as healthcare, finance, and robotics, where transparency and user trust are critical. Despite a lot of work done in the XAI field, most exist- ing methods fail to capture the dynamic nature of human-machine communication and have not been evaluated by human experiments. This thesis develops Dialogue Ex- planation Theory (DET), which incorporates interactive, conversational processes to deliver explanations, addressing how to achieve, improve and measure users’ compre- hension according to their individual requirements and understanding levels. Grounded in integrating this theory with rule-based reasoning systems, this research not only en- riches the intended user’s understanding of effective explanations in rule-based AI sys- tems but also extends its application to interpretable machine learning models. This approach has been implemented in rule-based reasoning AI systems from scratch and conducted several real-user studies. The findings suggest that DET can improve users’ understanding of the reasoning behind a conclusion by asking “Why?” and “Why not?” questions. It also gives machines the capacity to explain their actions, thereby enhanc- ing transparency and trust in such systems. This thesis underscores the importance of interaction in XAI and sets the groundwork for future explorations into user-focused AI explanations.

Date of Award	6 Jan 2025
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Louise Dennis (Supervisor), Joe Collenette (Supervisor) & Clare Dixon (Supervisor)